Collaborative_Filtering_from_scratch

Neflix_movie_recommendation

Implemented the approach by reading the research paper:

Empirical Analysis of Predictive Algorithms for Collaborative Filtering.

Approach:

1)I converted the dataset into a data frame format where rows represent users, columns represent movies and values represent ratings.
2)Fitted the dataset into Nearest Neighbor algorithm
3)Created a nn function which calculates 15 nearest neighbors of each user and then calculates the weight similarities between two user pairs at a time in a 2D matrix using the formula given in homework research paper.
4)Created a prediction function which takes userid and movieid as input and calculates predicted movie rating based on the formula in the paper.

Time taken to execute:

Took around 4 hours to find predictions of all test data points in Google collab, might take more time in other IDE and environment

Results:

RMSE on test (100478) data points: 0.9886979
MAE on test data points: 0.790226

Dataset:

Netflix movie dataset

How to Run:

For each file, please edit the Dataset path in the codes. Run each cell in the notebook file

Code files:

Please refer the ipynb files for the precomputed outputs of every code if you like.
1)collaborative_filtering.ipynb file:
This file contains the implementation of Collaborative Filtering as per the equation and concepts mentioned in the research paper given in the homework file.

• predictions.txt file contains predictions of ratings on test data